Fantasizing with Dual GPs in Bayesian Optimization and Active Learning

TL;DR

This paper introduces a sparse Dual GP approach that improves the efficiency of batch acquisition functions in Bayesian Optimization and Active Learning, enabling linear scaling and one-step updates for non-Gaussian likelihoods.

Contribution

It presents a novel sparse Dual GP parameterization that enhances computational efficiency and scalability for fantasizing batch acquisition functions.

Findings

Achieves linear scaling with batch size

Enables one-step updates for non-Gaussian likelihoods

Extends sparse models to greedy batch fantasizing acquisition functions

Abstract

Gaussian processes (GPs) are the main surrogate functions used for sequential modelling such as Bayesian Optimization and Active Learning. Their drawbacks are poor scaling with data and the need to run an optimization loop when using a non-Gaussian likelihood. In this paper, we focus on `fantasizing' batch acquisition functions that need the ability to condition on new fantasized data computationally efficiently. By using a sparse Dual GP parameterization, we gain linear scaling with batch size as well as one-step updates for non-Gaussian likelihoods, thus extending sparse models to greedy batch fantasizing acquisition functions.